NV-SRAM Data Backup Using Dead Block Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional caching structures based on static random access memory (SRAM) face limitations due to high leakage current and low integration degree, while direct use of nonvolatile memory (NVM) in caches leads to issues like high write power consumption, latency, and endurance, necessitating a more efficient data backup solution that reduces redundancy and improves system performance.
Innovation Solution
A data backup apparatus and method utilizing a nonvolatile static random access memory (NV-SRAM) with a backup power supply, where dead cache blocks are predicted using access sequence identifiers and counters to determine which blocks are live or dead, thereby reducing the volume of data to be backed up and improving efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of substance
If all data in SRAM is compressed and backed up to NVM, then data redundancy is reduced, but compression and decompression time increases affecting system performance
Solution Approach 1:
The patent extracts and identifies only the necessary data blocks (live blocks) that need to be backed up, rather than processing all data. By using access sequence identifiers and counters to detect dead blocks, the system extracts only the essential subset of data for backup, eliminating unnecessary compression operations on already-persisted data.
Solution Approach 2:
Instead of performing full compression on all SRAM data, the patent applies compression only to identified live blocks that require backup. This partial action approach reduces the scope of compression operations to only what is necessary, thereby reducing time overhead while still achieving redundancy reduction.
2Reliability
If a large on-chip energy storage capacitor is used to supply power for backing up data after power outage, then data backup capability is maintained, but chip area overheads and cost overheads increase
Solution Approach 1:
The patent performs partial backup by identifying and backing up only live blocks rather than all SRAM data. This reduces the total data volume requiring power supply during backup, thereby reducing the energy storage capacitor size needed while maintaining adequate backup capability for critical data.
Solution Approach 2:
The system performs preliminary identification of live blocks using access sequence identifiers and counters before the power outage occurs. This preliminary action enables the backup process to focus only on necessary data, reducing the power requirements and capacitor size needed during the actual backup operation.
3Reliability
If data is backed up in serial manner, then peak current is reduced, but backup time increases as backup data volume grows
Solution Approach 1:
By backing up only live blocks rather than all SRAM data, the patent reduces the total backup data volume. This allows the serial backup process to complete faster while maintaining system stability, as the reduced data set requires less total time even with sequential access.
Solution Approach 2:
The patent segments the backup process by identifying and separating live blocks from dead blocks. This segmentation enables the backup operation to process only relevant data segments, reducing overall backup time while the serial manner maintains system stability by avoiding peak current issues.
4Area of stationary object
If NVM is directly used as cache material, then integration degree and leakage current performance improve, but write power consumption, write latency, and write endurance deteriorate
Solution Approach 1:
The patent performs partial writes to NVM by backing up only live blocks rather than all cache data. This reduces the frequency and volume of write operations to NVM, thereby reducing write power consumption and wear on the NVM cells while maintaining the integration benefits of using NVM for cache.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present invention provides a data backup apparatus and method. According to the data backup method, when a data backup apparatus is powered on, a quantity of dead blocks and a quantity of live blocks are counted, where the dead blocks and the live blocks correspond to a sequence access identifier that is at a time point when the data backup apparatus is powered on. After the data backup apparatus is powered off, a proportion occupied by dead blocks corresponding to each sequence access identifier at the power-on time point in a total quantity of sampled cache blocks corresponding to the sequence access identifier, is calculated according to the counted quantities of dead blocks and live blocks that correspond to the sequence access identifier at the time point when the data backup apparatus is powered on. The calculated proportion is compared with a preset threshold, and a dead block in a volatile memory unit is predicted according to a comparison result. During backup, a cache block that is predicted to be a dead block is not backed up. This reduces a backup data volume and improves backup efficiency.